I am constantly amazed at the poor quality of data presented by many people trying to sell feed additives. Whilst it isn’t expected that every farmer should be a statistician, there are a few things one should keep an eye out for when presented with research data. To begin with, the source of the data is really important. Recognised Journals such as British Poultry Science are checked by other academics prior to publication and these are commonly referred to as “peer reviewed”. Peer reviewed publications carry a lot more weight than popular press articles or glossy presentations of “in house” research. Even large reputable companies are guilty of selective reporting if they are able to sell their products this way.
One of the cleverest tricks I see every day is where data is only reported as a percent improvement. For example, a simple comparison of the effects of a product on laying percentage in hens may report that product X improves the eggs per hen housed by 10% (which is a lot if production is already good). Problem here is that if one doesn’t know the base levels, there is no way that a figure such as this can be verified or checked against your current data.
The other common trick is to quote averages without statistics. An average of a set of data tells one absolutely nothing about the quality (or spread) of the data. If there are two treatments looking at body weight of birds and the “average” weight is 1 850g per bird. Group A may have a range of birds from 1 750 to 1 950g (200g range) while group B ranges from 1 650 to 2 050g (400g range). Group B is clearly more uneven and the next measure required is called standard deviation. All this measures is the average of the difference from the mean. Using our 2 examples of A and B above and assuming there are equivalent numbers (and weights) either side of the mean (i.e. a “normal distribution”) then the standard deviation or SD of A would be 50g and B would be double A at 100g. The coefficient of variation (or CV) is then simply a measure of SD expressed as a percentage of the mean. So A would have a CV of 2.7% and B would be 5.4%. Once we have a measure of variation, we are then able to analyse the data in a meaningful way to see if differences are real or simply part of the normal variation.
I will add more “statistics” to future issues of Out of the Bag but some simple explanations of statistics can be found at the following website: